Portable Stack Manipulation

This crate provides very portable functions to control the stack pointer and inspect the properties of the stack. This crate does not attempt to provide safe abstractions to any operations, the only goals are correctness, portability and efficiency (in that exact order). As a consequence most functions you’ll see in this crate are unsafe.

Unless you’re writing a safe abstraction over stack manipulation, this is not the crate you want. Instead consider one of the safe abstractions over this crate. A good place to look at is the crates.io’s reverse dependency list.

Platform support

The following table lists supported targets and architectures with notes on the level of current support and knowledge about the target. The three columns “Available”, “Tested” and “Callstack” imply an increasingly high level of support.

• “Available” basically means that the code builds and the assembly files have been written for the target;
• “Tested” means that the assembly code has been tested or otherwise verified to be correct. For most targets it also means that continuous integration is set up;
• “Callstack” means that the assembly code has been written with due care to support unwinding the stack and displaying the call frames (i.e. gdb backtrace works as expected).

Target has been tested locally.

Stacks allocated the usual way are not valid to be used on Windows and the functions to allocate a stack in a proper way is a Windows implementation detail. As a (unnecessarily slow and inflexible) alternative use Fibers.

The assembly files are available, but the target hasn’t been verified to build

The code technically works on my local machine, but exception handling does not correctly work on appveyor, which makes me unwilling to mark this as working yet.

Stacks allocated the usual way are not valid to be used on Windows and the functions to allocate a stack in a proper way is a Windows implementation detail. As a (unnecessarily slow and inflexible) alternative use Fibers.

aarch64-apple-ios has not been tested. iOS hardware is necessary to run these tests.

Stacks allocated the usual way are not valid to be used on Windows and the functions to allocate a stack in a proper way is a Windows implementation detail. As a (unnecessarily slow and inflexible) alternative use Fibers.

armv7-apple-ios has not been tested. iOS hardware is necessary to run these tests.

Stacks allocated the usual way are not valid to be used on Windows and the functions to allocate a stack in a proper way is a Windows implementation detail. As a (unnecessarily slow and inflexible) alternative use Fibers.

Only the o32 ABI is supported and will be used for all 32-bit MIPS targets.

Callstack generation may fail at certain well defined ranges of the program, although the usual compiler-generated code fails at similar points itself.

Callstack generation may fail at certain well defined ranges of the program, although the usual compiler-generated code fails at similar points itself.

Callstack generation may fail at certain well defined ranges of the program, although the usual compiler-generated code fails at similar points itself.

Test runner on CI hangs, local verification has been done on a qemu-system-s390x VM. It may be possible to add CI testing in the future via qemu’s full-system emulation.

A Rust target for 32-bit SPARC exists, but no Linux distributions actually have a 32-bit SPARC distribution, so verification is infeasible.

The actual assembly code has been written conservatively, modelled after the 64-bit SPARC code. and so has a non-zero chance of working.

Has been manually verified to work on the GCC Farm Project machines. It may be possible to add CI testing in the future via qemu’s full-system emulation.

Uses the same assembly as the sparc64-linux-gnu target. This target has no rustc builds and therefore the correct operation of this target could not be verified at the moment.

This library is not applicable to the target. WASM hasn’t a specified C ABI, the callstack is not even in an address space and does not appear to be manipulatable.

Feasibility/necessity hasn’t been acertained.

Feasibility/necessity hasn’t been acertained.

Haven’t gotten to it yet...

Although the assembly code has not been tested, it is a straightforward copy of the 64-bit version. Unless there is a non-obvious mistake, this should work fine.

The assembly code for riscv64 has been tested locally with a C caller.

The assembly code for loongarch64 has been tested locally with a C caller.

License

PSM is licensed under either of

• Apache License, Version 2.0, (LICENSE-APACHE or https://www.apache.org/licenses/LICENSE-2.0)
• MIT license (LICENSE-MIT or https://opensource.org/licenses/MIT)

at your option.